xref: /linux/arch/arm64/include/asm/atomic_lse.h (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
1 /*
2  * Based on arch/arm/include/asm/atomic.h
3  *
4  * Copyright (C) 1996 Russell King.
5  * Copyright (C) 2002 Deep Blue Solutions Ltd.
6  * Copyright (C) 2012 ARM Ltd.
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License version 2 as
10  * published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License
18  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
19  */
20 
21 #ifndef __ASM_ATOMIC_LSE_H
22 #define __ASM_ATOMIC_LSE_H
23 
24 #ifndef __ARM64_IN_ATOMIC_IMPL
25 #error "please don't include this file directly"
26 #endif
27 
28 #define __LL_SC_ATOMIC(op)	__LL_SC_CALL(atomic_##op)
29 
30 static inline void atomic_andnot(int i, atomic_t *v)
31 {
32 	register int w0 asm ("w0") = i;
33 	register atomic_t *x1 asm ("x1") = v;
34 
35 	asm volatile(ARM64_LSE_ATOMIC_INSN(__LL_SC_ATOMIC(andnot),
36 	"	stclr	%w[i], %[v]\n")
37 	: [i] "+r" (w0), [v] "+Q" (v->counter)
38 	: "r" (x1)
39 	: "x30");
40 }
41 
42 static inline void atomic_or(int i, atomic_t *v)
43 {
44 	register int w0 asm ("w0") = i;
45 	register atomic_t *x1 asm ("x1") = v;
46 
47 	asm volatile(ARM64_LSE_ATOMIC_INSN(__LL_SC_ATOMIC(or),
48 	"	stset	%w[i], %[v]\n")
49 	: [i] "+r" (w0), [v] "+Q" (v->counter)
50 	: "r" (x1)
51 	: "x30");
52 }
53 
54 static inline void atomic_xor(int i, atomic_t *v)
55 {
56 	register int w0 asm ("w0") = i;
57 	register atomic_t *x1 asm ("x1") = v;
58 
59 	asm volatile(ARM64_LSE_ATOMIC_INSN(__LL_SC_ATOMIC(xor),
60 	"	steor	%w[i], %[v]\n")
61 	: [i] "+r" (w0), [v] "+Q" (v->counter)
62 	: "r" (x1)
63 	: "x30");
64 }
65 
66 static inline void atomic_add(int i, atomic_t *v)
67 {
68 	register int w0 asm ("w0") = i;
69 	register atomic_t *x1 asm ("x1") = v;
70 
71 	asm volatile(ARM64_LSE_ATOMIC_INSN(__LL_SC_ATOMIC(add),
72 	"	stadd	%w[i], %[v]\n")
73 	: [i] "+r" (w0), [v] "+Q" (v->counter)
74 	: "r" (x1)
75 	: "x30");
76 }
77 
78 static inline int atomic_add_return(int i, atomic_t *v)
79 {
80 	register int w0 asm ("w0") = i;
81 	register atomic_t *x1 asm ("x1") = v;
82 
83 	asm volatile(ARM64_LSE_ATOMIC_INSN(
84 	/* LL/SC */
85 	"	nop\n"
86 	__LL_SC_ATOMIC(add_return),
87 	/* LSE atomics */
88 	"	ldaddal	%w[i], w30, %[v]\n"
89 	"	add	%w[i], %w[i], w30")
90 	: [i] "+r" (w0), [v] "+Q" (v->counter)
91 	: "r" (x1)
92 	: "x30", "memory");
93 
94 	return w0;
95 }
96 
97 static inline void atomic_and(int i, atomic_t *v)
98 {
99 	register int w0 asm ("w0") = i;
100 	register atomic_t *x1 asm ("x1") = v;
101 
102 	asm volatile(ARM64_LSE_ATOMIC_INSN(
103 	/* LL/SC */
104 	"	nop\n"
105 	__LL_SC_ATOMIC(and),
106 	/* LSE atomics */
107 	"	mvn	%w[i], %w[i]\n"
108 	"	stclr	%w[i], %[v]")
109 	: [i] "+r" (w0), [v] "+Q" (v->counter)
110 	: "r" (x1)
111 	: "x30");
112 }
113 
114 static inline void atomic_sub(int i, atomic_t *v)
115 {
116 	register int w0 asm ("w0") = i;
117 	register atomic_t *x1 asm ("x1") = v;
118 
119 	asm volatile(ARM64_LSE_ATOMIC_INSN(
120 	/* LL/SC */
121 	"	nop\n"
122 	__LL_SC_ATOMIC(sub),
123 	/* LSE atomics */
124 	"	neg	%w[i], %w[i]\n"
125 	"	stadd	%w[i], %[v]")
126 	: [i] "+r" (w0), [v] "+Q" (v->counter)
127 	: "r" (x1)
128 	: "x30");
129 }
130 
131 static inline int atomic_sub_return(int i, atomic_t *v)
132 {
133 	register int w0 asm ("w0") = i;
134 	register atomic_t *x1 asm ("x1") = v;
135 
136 	asm volatile(ARM64_LSE_ATOMIC_INSN(
137 	/* LL/SC */
138 	"	nop\n"
139 	__LL_SC_ATOMIC(sub_return)
140 	"	nop",
141 	/* LSE atomics */
142 	"	neg	%w[i], %w[i]\n"
143 	"	ldaddal	%w[i], w30, %[v]\n"
144 	"	add	%w[i], %w[i], w30")
145 	: [i] "+r" (w0), [v] "+Q" (v->counter)
146 	: "r" (x1)
147 	: "x30", "memory");
148 
149 	return w0;
150 }
151 
152 #undef __LL_SC_ATOMIC
153 
154 #define __LL_SC_ATOMIC64(op)	__LL_SC_CALL(atomic64_##op)
155 
156 static inline void atomic64_andnot(long i, atomic64_t *v)
157 {
158 	register long x0 asm ("x0") = i;
159 	register atomic64_t *x1 asm ("x1") = v;
160 
161 	asm volatile(ARM64_LSE_ATOMIC_INSN(__LL_SC_ATOMIC64(andnot),
162 	"	stclr	%[i], %[v]\n")
163 	: [i] "+r" (x0), [v] "+Q" (v->counter)
164 	: "r" (x1)
165 	: "x30");
166 }
167 
168 static inline void atomic64_or(long i, atomic64_t *v)
169 {
170 	register long x0 asm ("x0") = i;
171 	register atomic64_t *x1 asm ("x1") = v;
172 
173 	asm volatile(ARM64_LSE_ATOMIC_INSN(__LL_SC_ATOMIC64(or),
174 	"	stset	%[i], %[v]\n")
175 	: [i] "+r" (x0), [v] "+Q" (v->counter)
176 	: "r" (x1)
177 	: "x30");
178 }
179 
180 static inline void atomic64_xor(long i, atomic64_t *v)
181 {
182 	register long x0 asm ("x0") = i;
183 	register atomic64_t *x1 asm ("x1") = v;
184 
185 	asm volatile(ARM64_LSE_ATOMIC_INSN(__LL_SC_ATOMIC64(xor),
186 	"	steor	%[i], %[v]\n")
187 	: [i] "+r" (x0), [v] "+Q" (v->counter)
188 	: "r" (x1)
189 	: "x30");
190 }
191 
192 static inline void atomic64_add(long i, atomic64_t *v)
193 {
194 	register long x0 asm ("x0") = i;
195 	register atomic64_t *x1 asm ("x1") = v;
196 
197 	asm volatile(ARM64_LSE_ATOMIC_INSN(__LL_SC_ATOMIC64(add),
198 	"	stadd	%[i], %[v]\n")
199 	: [i] "+r" (x0), [v] "+Q" (v->counter)
200 	: "r" (x1)
201 	: "x30");
202 }
203 
204 static inline long atomic64_add_return(long i, atomic64_t *v)
205 {
206 	register long x0 asm ("x0") = i;
207 	register atomic64_t *x1 asm ("x1") = v;
208 
209 	asm volatile(ARM64_LSE_ATOMIC_INSN(
210 	/* LL/SC */
211 	"	nop\n"
212 	__LL_SC_ATOMIC64(add_return),
213 	/* LSE atomics */
214 	"	ldaddal	%[i], x30, %[v]\n"
215 	"	add	%[i], %[i], x30")
216 	: [i] "+r" (x0), [v] "+Q" (v->counter)
217 	: "r" (x1)
218 	: "x30", "memory");
219 
220 	return x0;
221 }
222 
223 static inline void atomic64_and(long i, atomic64_t *v)
224 {
225 	register long x0 asm ("x0") = i;
226 	register atomic64_t *x1 asm ("x1") = v;
227 
228 	asm volatile(ARM64_LSE_ATOMIC_INSN(
229 	/* LL/SC */
230 	"	nop\n"
231 	__LL_SC_ATOMIC64(and),
232 	/* LSE atomics */
233 	"	mvn	%[i], %[i]\n"
234 	"	stclr	%[i], %[v]")
235 	: [i] "+r" (x0), [v] "+Q" (v->counter)
236 	: "r" (x1)
237 	: "x30");
238 }
239 
240 static inline void atomic64_sub(long i, atomic64_t *v)
241 {
242 	register long x0 asm ("x0") = i;
243 	register atomic64_t *x1 asm ("x1") = v;
244 
245 	asm volatile(ARM64_LSE_ATOMIC_INSN(
246 	/* LL/SC */
247 	"	nop\n"
248 	__LL_SC_ATOMIC64(sub),
249 	/* LSE atomics */
250 	"	neg	%[i], %[i]\n"
251 	"	stadd	%[i], %[v]")
252 	: [i] "+r" (x0), [v] "+Q" (v->counter)
253 	: "r" (x1)
254 	: "x30");
255 }
256 
257 static inline long atomic64_sub_return(long i, atomic64_t *v)
258 {
259 	register long x0 asm ("x0") = i;
260 	register atomic64_t *x1 asm ("x1") = v;
261 
262 	asm volatile(ARM64_LSE_ATOMIC_INSN(
263 	/* LL/SC */
264 	"	nop\n"
265 	__LL_SC_ATOMIC64(sub_return)
266 	"	nop",
267 	/* LSE atomics */
268 	"	neg	%[i], %[i]\n"
269 	"	ldaddal	%[i], x30, %[v]\n"
270 	"	add	%[i], %[i], x30")
271 	: [i] "+r" (x0), [v] "+Q" (v->counter)
272 	: "r" (x1)
273 	: "x30", "memory");
274 
275 	return x0;
276 }
277 
278 static inline long atomic64_dec_if_positive(atomic64_t *v)
279 {
280 	register long x0 asm ("x0") = (long)v;
281 
282 	asm volatile(ARM64_LSE_ATOMIC_INSN(
283 	/* LL/SC */
284 	"	nop\n"
285 	__LL_SC_ATOMIC64(dec_if_positive)
286 	"	nop\n"
287 	"	nop\n"
288 	"	nop\n"
289 	"	nop\n"
290 	"	nop",
291 	/* LSE atomics */
292 	"1:	ldr	x30, %[v]\n"
293 	"	subs	%[ret], x30, #1\n"
294 	"	b.lt	2f\n"
295 	"	casal	x30, %[ret], %[v]\n"
296 	"	sub	x30, x30, #1\n"
297 	"	sub	x30, x30, %[ret]\n"
298 	"	cbnz	x30, 1b\n"
299 	"2:")
300 	: [ret] "+&r" (x0), [v] "+Q" (v->counter)
301 	:
302 	: "x30", "cc", "memory");
303 
304 	return x0;
305 }
306 
307 #undef __LL_SC_ATOMIC64
308 
309 #define __LL_SC_CMPXCHG(op)	__LL_SC_CALL(__cmpxchg_case_##op)
310 
311 #define __CMPXCHG_CASE(w, sz, name, mb, cl...)				\
312 static inline unsigned long __cmpxchg_case_##name(volatile void *ptr,	\
313 						  unsigned long old,	\
314 						  unsigned long new)	\
315 {									\
316 	register unsigned long x0 asm ("x0") = (unsigned long)ptr;	\
317 	register unsigned long x1 asm ("x1") = old;			\
318 	register unsigned long x2 asm ("x2") = new;			\
319 									\
320 	asm volatile(ARM64_LSE_ATOMIC_INSN(				\
321 	/* LL/SC */							\
322 	"	nop\n"							\
323 		__LL_SC_CMPXCHG(name)					\
324 	"	nop",							\
325 	/* LSE atomics */						\
326 	"	mov	" #w "30, %" #w "[old]\n"			\
327 	"	cas" #mb #sz "\t" #w "30, %" #w "[new], %[v]\n"		\
328 	"	mov	%" #w "[ret], " #w "30")			\
329 	: [ret] "+r" (x0), [v] "+Q" (*(unsigned long *)ptr)		\
330 	: [old] "r" (x1), [new] "r" (x2)				\
331 	: "x30" , ##cl);						\
332 									\
333 	return x0;							\
334 }
335 
336 __CMPXCHG_CASE(w, b,    1,   )
337 __CMPXCHG_CASE(w, h,    2,   )
338 __CMPXCHG_CASE(w,  ,    4,   )
339 __CMPXCHG_CASE(x,  ,    8,   )
340 __CMPXCHG_CASE(w, b, mb_1, al, "memory")
341 __CMPXCHG_CASE(w, h, mb_2, al, "memory")
342 __CMPXCHG_CASE(w,  , mb_4, al, "memory")
343 __CMPXCHG_CASE(x,  , mb_8, al, "memory")
344 
345 #undef __LL_SC_CMPXCHG
346 #undef __CMPXCHG_CASE
347 
348 #define __LL_SC_CMPXCHG_DBL(op)	__LL_SC_CALL(__cmpxchg_double##op)
349 
350 #define __CMPXCHG_DBL(name, mb, cl...)					\
351 static inline int __cmpxchg_double##name(unsigned long old1,		\
352 					 unsigned long old2,		\
353 					 unsigned long new1,		\
354 					 unsigned long new2,		\
355 					 volatile void *ptr)		\
356 {									\
357 	unsigned long oldval1 = old1;					\
358 	unsigned long oldval2 = old2;					\
359 	register unsigned long x0 asm ("x0") = old1;			\
360 	register unsigned long x1 asm ("x1") = old2;			\
361 	register unsigned long x2 asm ("x2") = new1;			\
362 	register unsigned long x3 asm ("x3") = new2;			\
363 	register unsigned long x4 asm ("x4") = (unsigned long)ptr;	\
364 									\
365 	asm volatile(ARM64_LSE_ATOMIC_INSN(				\
366 	/* LL/SC */							\
367 	"	nop\n"							\
368 	"	nop\n"							\
369 	"	nop\n"							\
370 	__LL_SC_CMPXCHG_DBL(name),					\
371 	/* LSE atomics */						\
372 	"	casp" #mb "\t%[old1], %[old2], %[new1], %[new2], %[v]\n"\
373 	"	eor	%[old1], %[old1], %[oldval1]\n"			\
374 	"	eor	%[old2], %[old2], %[oldval2]\n"			\
375 	"	orr	%[old1], %[old1], %[old2]")			\
376 	: [old1] "+r" (x0), [old2] "+r" (x1),				\
377 	  [v] "+Q" (*(unsigned long *)ptr)				\
378 	: [new1] "r" (x2), [new2] "r" (x3), [ptr] "r" (x4),		\
379 	  [oldval1] "r" (oldval1), [oldval2] "r" (oldval2)		\
380 	: "x30" , ##cl);						\
381 									\
382 	return x0;							\
383 }
384 
385 __CMPXCHG_DBL(   ,   )
386 __CMPXCHG_DBL(_mb, al, "memory")
387 
388 #undef __LL_SC_CMPXCHG_DBL
389 #undef __CMPXCHG_DBL
390 
391 #endif	/* __ASM_ATOMIC_LSE_H */
392